Skip to content

Vulkan MMQ Fix #8479

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jul 15, 2024
Merged

Vulkan MMQ Fix #8479

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
18b2ac1
Fix incoherence by adding missing LOAD_VEC_A parameter
0cc4m Jul 14, 2024
cc7a2f6
Fix Vulkan op result checker build error
0cc4m Jul 14, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions ggml/src/ggml-vulkan.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6561,7 +6561,7 @@ static void ggml_vk_print_tensor(ggml_backend_vk_context * ctx, const ggml_tenso
ggml_tensor_extra_gpu * extra = (ggml_tensor_extra_gpu *) tensor->extra;

vk_buffer buffer_gpu = extra->buffer_gpu.lock();
ggml_vk_buffer_read(ctx, buffer_gpu, extra->offset + tensor->view_offs, tensor_data, tensor_size);
ggml_vk_buffer_read(buffer_gpu, extra->offset + tensor->view_offs, tensor_data, tensor_size);
}

std::cerr << "TENSOR CHECK " << name << " (" << tensor->name << "): " << ggml_op_name(tensor->op) << std::endl;
Expand Down Expand Up @@ -6645,7 +6645,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
for (int i3 = 0; i3 < src0->ne[3]; i3++) {
for (int i2 = 0; i2 < src0->ne[2]; i2++) {
const int idx = i3*src0->ne[2] + i2;
ggml_vk_buffer_read(ctx, buffer_gpu, offset + idx * src0->nb[2], ((char *)src0_clone->data + idx * src0_clone->nb[2]), src0->ne[1] * src0->nb[1]);
ggml_vk_buffer_read(buffer_gpu, offset + idx * src0->nb[2], ((char *)src0_clone->data + idx * src0_clone->nb[2]), src0->ne[1] * src0->nb[1]);
}
}

Expand All @@ -6658,7 +6658,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
if (offset + src0_size >= buffer_gpu->size) {
src0_size = buffer_gpu->size - offset;
}
ggml_vk_buffer_read(ctx, buffer_gpu, offset, src0_clone->data, src0_size);
ggml_vk_buffer_read(buffer_gpu, offset, src0_clone->data, src0_size);
memcpy(src0_clone->nb, src0->nb, sizeof(size_t) * GGML_MAX_DIMS);
}
} else {
Expand Down Expand Up @@ -6687,7 +6687,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
for (int i3 = 0; i3 < src1->ne[3]; i3++) {
for (int i2 = 0; i2 < src1->ne[2]; i2++) {
const int idx = i3*src1->ne[2] + i2;
ggml_vk_buffer_read(ctx, buffer_gpu, offset + idx * src1->nb[2], ((char *)src1_clone->data + idx * src1_clone->nb[2]), src1->ne[1] * src1->nb[1]);
ggml_vk_buffer_read(buffer_gpu, offset + idx * src1->nb[2], ((char *)src1_clone->data + idx * src1_clone->nb[2]), src1->ne[1] * src1->nb[1]);
}
}

Expand All @@ -6700,7 +6700,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
if (offset + src1_size >= buffer_gpu->size) {
src1_size = buffer_gpu->size - offset;
}
ggml_vk_buffer_read(ctx, buffer_gpu, offset, src1_clone->data, src1_size);
ggml_vk_buffer_read(buffer_gpu, offset, src1_clone->data, src1_size);
memcpy(src1_clone->nb, src1->nb, sizeof(size_t) * GGML_MAX_DIMS);
}
} else {
Expand Down Expand Up @@ -6745,7 +6745,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
for (int i3 = 0; i3 < src2->ne[3]; i3++) {
for (int i2 = 0; i2 < src2->ne[2]; i2++) {
const int idx = i3*src2->ne[2] + i2;
ggml_vk_buffer_read(ctx, buffer_gpu, offset + idx * src2->nb[2], ((char *)src2_clone->data + idx * src2_clone->nb[2]), src2->ne[1] * src2->nb[1]);
ggml_vk_buffer_read(buffer_gpu, offset + idx * src2->nb[2], ((char *)src2_clone->data + idx * src2_clone->nb[2]), src2->ne[1] * src2->nb[1]);
}
}

Expand All @@ -6758,7 +6758,7 @@ static void ggml_vk_check_results_0(ggml_backend_vk_context * ctx, ggml_tensor *
if (offset + src2_size >= buffer_gpu->size) {
src2_size = buffer_gpu->size - offset;
}
ggml_vk_buffer_read(ctx, buffer_gpu, offset, src2_clone->data, src2_size);
ggml_vk_buffer_read(buffer_gpu, offset, src2_clone->data, src2_size);
memcpy(src2_clone->nb, src2->nb, sizeof(size_t) * GGML_MAX_DIMS);
}
} else {
Expand Down Expand Up @@ -6922,7 +6922,7 @@ static void ggml_vk_check_results_1(ggml_backend_vk_context * ctx, ggml_tensor *
tensor_size = buffer_gpu->size - (extra->offset + tensor->view_offs);
}

ggml_vk_buffer_read(ctx, buffer_gpu, extra->offset + tensor->view_offs, tensor_data, tensor_size);
ggml_vk_buffer_read(buffer_gpu, extra->offset + tensor->view_offs, tensor_data, tensor_size);
}

float first_error_result = -1.0f;
Expand Down
4 changes: 2 additions & 2 deletions ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -270,10 +270,10 @@ void matmul_shaders(std::vector<std::future<void>>& tasks, bool fp16, bool matmu
std::string data_a_key = "DATA_A_" + to_uppercase(tname);
std::string load_vec_a = (tname == "f32" || tname == "f16") ? load_vec : "2";
tasks.push_back(std::async(std::launch::async, [=] {
string_to_spv(shader_name + "_" + tname + "_f32", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16);
string_to_spv(shader_name + "_" + tname + "_f32", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"B_TYPE", "float"}, {"D_TYPE", "float"}}), fp16);
}));
tasks.push_back(std::async(std::launch::async, [=] {
string_to_spv(shader_name + "_" + tname + "_f32_aligned", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "2"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}}), fp16);
string_to_spv(shader_name + "_" + tname + "_f32_aligned", "mul_mm.comp", merge_maps(base_dict, {{data_a_key, "1"}, {"LOAD_VEC_A", load_vec_a}, {"LOAD_VEC_B", load_vec}, {"B_TYPE", aligned_b_type_f32}, {"D_TYPE", "float"}}), fp16);
}));
}
}
Expand Down
Loading